Optimizing the Collaborative Supply Scheduling of Renewable Energy in the Tourism Industry Chain Using Graph Neural Networks

ABSTRACT To cope with the supply and demand imbalance challenge caused by the surge in energy demand in tourism‐intensive areas and the high volatility of renewable energy, this paper aims to construct a collaborative scheduling optimization model based on graph neural networks (GNNs). The model abstracts load nodes with spatiotemporal correlation and energy complementarity characteristics such as hotels, transportation hubs, and scenic spots in the tourism industry chain and energy facilities into a dynamic heterogeneous graph structure, and precisely represents the spatiotemporal coupling relationship between multiple subjects through node embedding and neighborhood aggregation. To address the voltage over‐limit and abandoned light problems caused by the intermittent nature of renewable energy photovoltaics, a GNN‐driven “prediction‐cooperative scheduling” joint strategy is designed in combination with a multi‐time‐scale scheduling framework. In the day‐ahead stage, Graph Convolutional Networks (GCNs) are used to fuse historical load and meteorological data to generate high‐precision wind and solar power output forecasts. In the real‐time stage, the node status and association strength are dynamically perceived based on the graph attention network (GAT); the energy storage charging and discharging power is adaptively allocated; the scheduling strategy is trained through reinforcement learning (RL) to collaboratively optimize energy costs and carbon emission targets. The empirical results show that, based on the actual data verification of the Anhui “Photovoltaic + Tourism” demonstration area, in the collaborative scheduling of the Huangshan Scenic Area microgrid group, the model increases the clean energy consumption rate to 98.2%, which is 9.1% higher than the model predictive control MPC strategy; the voltage over‐limit duration is significantly reduced by 87.5% (from 3.2 to 0.4 h); the penalty cost of abandoned light is sharply reduced by 92%; the response speed is 46% higher than the PSO (Particle Swarm Optimization) algorithm; the total operating cost is reduced by 15.5%. This paper provides a scalable technical path for the construction of an efficient self‐balancing system in the tourism ecosystem, which significantly helps the high‐quality development of low‐carbon tourism.